The state will be unavailable. If you want to trigger from on to off or from on to unavailable or unknown do this:
trigger:
- platform: state
entity_id: light.my_smart_light
from: 'on'
if you only want to trigger from on to off do:
trigger:
- platform: state
entity_id: light.my_smart_light
from: 'on'
to: 'off'
Correct. Unavailable is unavailable, unknown is unknown, neither would be considered as off.
Off is off.
I had 1 of those, mainly as an experiment in the UI editor for automations (I have them all in yaml.), and it was even more cryptical than most above:
- type: moist
platform: device
device_id: afa7a35223f4392eaaf6c1b88ab072db
entity_id: 2adfe81af765e65dad7ce32ca8bdf1e9
domain: binary_sensor
I mean, we can deduce its about a sensor detecting moist, but it doesnt even contain a to state? only a type?
- platform: state
entity_id: binary_sensor.waterlek_sensor
to: 'on'
# from: 'off' # if you require so
yep definitely agree on the easier choice here…
funny thing is that the automation fires a notification, during the period the device is ‘on’, and that actually does:
while:
- condition: state
entity_id: binary_sensor.waterlek_sensor
state: 'on'
so even the UI thinks its better to use that entity_id 
If you want to rename the device and all the entities below are modified, then you have to go to all the automation to modifying those.
Using device_id will not be impacted by this. There are some pro and cons.
That was helpful, for me at least, in understanding how that works.
And in turn it has me appreciate the Watchman integration even more, and thought the integration deserved a mention in for those even greener than I am with HA, in this regard.
Watchman is a great help in there.
Also there other benefits to use device_id as this will be used by HA in some parts of the code like in logbook apparently (I’m just repeating info that I got from the forum).
Check my Feature request if you are interested:
When you onboard a new device you should first give it a sensible name. If you’re renaming after you’ve started using it, that’s a different matter, but you likely will only have to do the search and replace once ever (assuming it gets a proper name the second time around). If you replace a device you will have to rename the usages every time you replace a device, as opposed to renaming device entities.
Renaming entities is a choice, if you do not, automations will work, if you do change them, it is exactly the same amount of work as changing device IDs.
The biggest dealbreakers using device ids for me are:
How to avoid device_id for scene controller?
I have a Z-Wave button, or really a “scene controller”, that I use to turn on a pump. If I use the UI to create an automation for scene 002:
The resulting action looks like this, which does trigger:
trigger:
- platform: device
device_id: f381791969ec48f1a1405bfe1f9f39b0
domain: zwave_js
type: event.value_notification.central_scene
property: scene
property_key: "002"
endpoint: 0
command_class: 91
subtype: Endpoint 0 Scene 002
Here’s the entity for that scene:
I can test state like this, but that only triggers on the attribute changing – yes, it will trigger if I first press the button 2x times to set KeyPressed2x, but if it is already KeyPressed it won’t trigger since it’s not changing.
trigger:
- platform: state
entity_id:
- event.recirc_pump_remote_switch_scene_002
attribute: event_type
to: KeyPressed
I tried using an event trigger like this, but it’s not triggering. Plus, I’m wondering if triggering on such a frequent event is a bad idea.
trigger:
- platform: event
event_type: state_changed
event_data:
new_state:
entity_id: event.recirc_pump_remote_switch_scene_002
attributes:
event_type: KeyPressed
Is there another approach?
Here’s what the event looks like:
event_type: state_changed
data:
entity_id: event.recirc_pump_remote_switch_scene_002
old_state:
entity_id: event.recirc_pump_remote_switch_scene_002
state: "2024-06-09T17:39:03.528+00:00"
attributes:
event_types:
- KeyHeldDown
- KeyPressed
- KeyPressed2x
- KeyPressed3x
- KeyPressed4x
- KeyPressed5x
- KeyReleased
event_type: KeyPressed
value: 0
friendly_name: Recirc Pump Remote Switch Scene 002
last_changed: "2024-06-09T17:39:03.528460+00:00"
last_reported: "2024-06-09T17:39:03.528460+00:00"
last_updated: "2024-06-09T17:39:03.528460+00:00"
context:
id: 01HZZ1A3784S14M58679N86CM9
parent_id: null
user_id: null
new_state:
entity_id: event.recirc_pump_remote_switch_scene_002
state: "2024-06-09T17:47:41.061+00:00"
attributes:
event_types:
- KeyHeldDown
- KeyPressed
- KeyPressed2x
- KeyPressed3x
- KeyPressed4x
- KeyPressed5x
- KeyReleased
event_type: KeyPressed
value: 0
friendly_name: Recirc Pump Remote Switch Scene 002
last_changed: "2024-06-09T17:47:41.061750+00:00"
last_reported: "2024-06-09T17:47:41.061750+00:00"
last_updated: "2024-06-09T17:47:41.061750+00:00"
context:
id: 01HZZ1SWM57FP2QKQH328A7YWF
parent_id: null
user_id: null
origin: LOCAL
time_fired: "2024-06-09T17:47:41.061750+00:00"
context:
id: 01HZZ1SWM57FP2QKQH328A7YWF
parent_id: null
user_id: null
trigger:
- platform: state
entity_id: event.recirc_pump_remote_switch_scene_002
to: null
condition:
- condition: state
entity_id: event.recirc_pump_remote_switch_scene_002
attribute: event_type
state: KeyPressed
this would be a solution.
But I would combine all actions for each event type in one automation using choose
The most usual reason for me to change a name is that I am repurposing the device, so I actually find it useful that it breaks stuff noticeably when I rename - as of course I need to edit those to accept the replacement device. I tried leaving the entity names as they were but I find it impossible to recognise which one to use unless they bear the name of their parent device - as it has become.
I think this post should not be pinned. It’s a good post, but not important enough for all users to follow.
Disclaimer: I’m not saying anything in the guide is wrong (since the author obviously knows his stuff). I’m only challenging the relevance. Maybe this discussion will lead to firther improvements to the article or at worst, I learn that my assumptions are wring. I’ll take a bit of an oppositie opinion here and there with the expectation to end up in the middle.
Why this post should not be pinned (opinions)
Let’s go through the arguments:
The yaml will be longer and harder to follow…
This is only relevant if you use yaml primarily. For UI users, using a device is often easier to use and follow.
Looking at the majority of users, +1 for “using devices”.
If you replace a device you will have to go through every automation it appears in…
True, but the same can be said in favor of devices to avoid updating when renaming an entity (which I do more often than replacing devices)
I can’t say which scenario is more common in general, so let’s call it a tie.
Automations using device_id report as unavailable if they depend on missing devices.
This could be an important argument, but the impact is not explained. Will the automation not work?
It’s inconclusive for now, but could be a + for “not using devices”.
Using entity states you can exclude “unavailable” and “unknown” from your triggers:
This is a good suggestion to make automations more robust, but how many people actually use it? Beginners should know when to use it, but in a lot of cases, it won’t be relevant.
+1 for knowing “when to use devices and when to use entities”.
Device actions, triggers and conditions do not support more advanced features like templating.
As.abive, it’s not a reason not to use devices, but it might be a reason to know “when to use devices and when to use entities”.
To summarize:
Until it breaks down and needs to be replaced by a new one…
I disagree. It’s worth the pin.
You need to consider the other side of the coin too: Reading YAML and helping people using device IDs are harder — and I say this as an experienced user.
“Help us help you.”
It really isn’t that hard to grasp the concept of entities being the core concept and that devices are basically a grouping of entities.
It’s such a foundational cencept that it is unlikely to change ever, I think. It’s better to make the shift earlier before having to rewrite lots of things later.
Very much worth the pin.
Almost all of the problems I’ve had with automations has stemmed from using devices early on in my HASS experience. Things stopped working without any notification. All because I changed batteries on a few sensors and the new devices (obviously) got new device_ids.
How did changing batteries cause issues with automations exactly?
Perhaps an issue with the specific integration creating new device IDs?
Either way, this is a real problem when people do replace devices, say a blown smart bulb with a new one: It will be a new device, compared to just renaming and entity and you’re done.
This is all fine and good, until you use Zigbee switches (buttons, not relays) in ZHA. They don’t have entities except battery etc. So you are stuck on using event triggers and you have the choice of either device_id, unique_id or device_ieee.
That type of issue will likely be addressed by the use of event entities if/when they get added into ZHA. Like a lot of new features in HA, it was discussed and implemented; but it takes time and resources to add those features to each of the 2000+ core integrations as needed.
IIRC, ZHA has already been undergoing some significant updates, so it likely the case that those are taking priority over adding event entities.